import pandas as pd
import numpy as np
import math
import json
import pickle

import plotly.graph_objects as go
from datetime import date
from tqdm import tqdm

# read in the json files
portfolio = pd.read_json('Starbucks_Offers/data/portfolio.json', orient='records', lines=True)
profile = pd.read_json('Starbucks_Offers/data/profile.json', orient='records', lines=True)
transcript = pd.read_json('Starbucks_Offers/data/transcript.json', orient='records', lines=True)

profile.head()

sum(pd.isnull(profile['income']))

2175

profile = profile.dropna()

profile['became_member_on'] = profile['became_member_on'].apply(lambda x: '/'.join([str(x)[:4], str(x)[4:6], str(x)[6:]]))

profile['became_member_on'] = pd.to_datetime(profile['became_member_on'])

profile.head()

transcript['event'].unique()

array(['offer received', 'offer viewed', 'transaction', 'offer completed'],
      dtype=object)

transcript['value'] = transcript['value'].apply(lambda x: list(x.values())[0])

transcript.head()

transcript.shape

(306534, 4)

np.sum(transcript.duplicated())

397

transcript = transcript.drop_duplicates()

transcript['personvalue'] = transcript['person'].astype(str) + '-' + transcript['value'].astype(str) 

transcript.head()

What is the type of offer that leads most to transaction?¶

An offer which leads to a transaction is an offer that has been first viewed and then completed.

It can happen that an offer is being completed without the offer being viewed or before the offer being viewed.

For each person, we will then look at

time_viewed: the first time when the person has viewed the author

time_completed: the time when this person has completed the offer (the highest time if the offer hase been completed several times)

I time_completed > time_viewed, then we can consider that this offer has effectively led to a transaction.

As the following cell takes time running, we will comment it after having saved the output in a pickle file.

# offer_viewed = transcript[transcript['event'] == 'offer viewed']['personvalue']

# offer_viewed_and_completed = transcript[(transcript['personvalue'].isin(offer_viewed))
#                                        & (transcript['event'] == 'offer completed')]

# offers_leading_to_transaction = []
# edge_cases = []
# for offer in tqdm(offer_viewed_and_completed.sort_values('person')['personvalue'].unique()):
    
#     # focus on the transcript for one offer for a user
#     offer_timing = transcript[transcript['personvalue'] == offer].sort_values('time')
    
#     # selecting the first time the offer was viewed
#     time_viewed = offer_timing[offer_timing['event'] == 'offer viewed']['time'].iloc[0]
    
#     # selecting the last time the offer was completed
#     time_completed = offer_timing[offer_timing['event'] == 'offer completed']['time'].iloc[-1]
    
#     if time_completed >= time_viewed:
#         offers_leading_to_transaction.append(offer)
#     else:
#         edge_cases.append(offer)

# pickle.dump(offers_leading_to_transaction, open('Starbucks_Offers/offer_leading_to_transaction', 'wb'))
# pickle.dump(edge_cases, open('Starbucks_Offers/edge_cases', 'wb'))

offers_leading_to_transaction = pickle.load(open('Starbucks_Offers/offer_leading_to_transaction', 'rb'))
edge_cases = pickle.load(open('Starbucks_Offers/edge_cases', 'rb'))

Out of the offer received, percentage of offers that led effectively led to a transaction:

len(offers_leading_to_transaction)/len(transcript[transcript['event'] == 'offer received'])

0.28018931001481445

Out of the offer completed and viewed, percentage of offers viewed after having been completed

len(edge_cases)/(len(offers_leading_to_transaction) + len(edge_cases))

0.12624693376941945

types_leading_to_buy = transcript[
    transcript['personvalue'].isin(offers_leading_to_transaction)
]['value'].value_counts(normalize=True).reset_index()

types_leading_to_buy.columns = ['id', 'share_of_transactions']

types_leading_to_buy = portfolio.merge(types_leading_to_buy,
                                       on='id',
                                       how='outer').sort_values(by='share_of_transactions',
                                                                ascending=False)

overview_types = types_leading_to_buy.groupby('offer_type')['share_of_transactions'].sum()

fig = go.Figure()

fig.add_trace(go.Pie(labels=overview_types.index, 
                     values=overview_types))

fig.update_layout({'width':500, 'height':500})

fig.show()

Overall, BOGO offers and discounts are responsible for the same number of transactions.

types_leading_to_buy

What we can see here is that the discount offers of low difficulty and shared on all the channels, including social, tend to lead more to transactions than other discount offer that are more difficult to complete and shared through less channels.

Between them, are the BOGO offer. It seems that a BOGO offer will work better if the difficulty is low and if it is shared on social.

It is quite interesting to note that the reward does not make an offer more attractive. What seems to impact the most is the amount the person need to spend to get it as well as the place where the offer was shared.

For the following of the study, we will name each offer for more convenience. We will use the ranking of each offer and their type to name them.

We will leave the informational as they are because they won't be part of the offers leading to transaction.

types_leading_to_buy['offer_name'] = [
    'discount_1', 
    'discount_2', 
    'bogo_1', 
    'bogo_2', 
    'bogo_3',
    'bogo_4', 
    'discount_3', 
    'discount_4', 
    'informational', 
    'informational'
]

What is the type of offers generating the more revenue?¶

transaction_offer_completed = transcript[
    (transcript['personvalue'].isin(offers_leading_to_transaction))
    &(transcript['event'] == 'offer completed')][['person', 'time', 'value']
                                                ]

transaction_offer_completed.head()

transactions = transcript[transcript['event'] == 'transaction'][['person', 'time', 'value']]

revenue_per_offer = transaction_offer_completed.merge(transactions, how='left', left_on=['person','time'], right_on = ['person','time'])

revenue_per_offer.head()

revenue_per_type = revenue_per_offer.groupby('value_x')['value_y'].sum().reset_index()

revenue_per_type.columns = ['id', 'revenue']
revenue_per_type['revenue_share'] = revenue_per_type['revenue'] / revenue_per_type['revenue'].sum()

revenue_per_type = revenue_per_type.drop('revenue', axis=1)


types_leading_to_buy.merge(revenue_per_type,
                            on='id',
                            how='outer').sort_values('revenue_share', ascending=False)

The revenue led by each offers follows the same ranking as the number of transactions which means that an offer might not lead to spend always more than another one.

What is the profile of the persons completing the offers¶

focus_transcript = transcript[(transcript['personvalue'].isin(offers_leading_to_transaction))
                             & (transcript['event'] == 'offer completed')]

sum(focus_transcript[['person', 'value']].duplicated())

3866

There are 3866 times were the same person uses the offer several times. We will leave these duplicated lines as they provide us with the info of a specific offer working on certain types of person (even though we hope at the end winning this consumer as somebody that buys without any offer).

offers_and_persons = focus_transcript[['person', 'value']].merge(types_leading_to_buy[['id', 'offer_name']],
                                                                 how='left',
                                                                 left_on='value',
                                                                 right_on='id')

offers_and_persons = offers_and_persons.drop(['id', 'value'], axis=1)
offers_and_persons

Now let's use the profiling of users to understand better which user will use which offer.

offers_and_persons = offers_and_persons.merge(profile, how='left', left_on='person', right_on='id')
offers_and_persons

1. The gender¶

sum(pd.isnull(offers_and_persons['gender']))

974

# Removing Nan values

offers_and_persons = offers_and_persons[~pd.isnull(offers_and_persons['gender'])]

gender_overall = offers_and_persons['gender'].value_counts(normalize=True)

fig = go.Figure()

fig.add_trace(go.Pie(values=gender_overall, 
                     labels=gender_overall.index))

fig.update_layout(width=500, height=500)

fig.show()

In the transactions we are looking at, we have almost the same number of males and females. Let's look at this stat per offer.

gender_per_offer = offers_and_persons.groupby(['offer_name', 'gender'])['id'].count().reset_index()

gender_per_offer = gender_per_offer.merge(gender_per_offer.groupby('offer_name')['id'].sum(),
                                          how='left',
                                          on='offer_name')

gender_per_offer['id_x'] = gender_per_offer['id_x']/gender_per_offer['id_y']

gender_per_offer.head()

fig = go.Figure()
for gender in ['F', 'M', 'O']:    
            
    fig.add_trace(go.Bar(
        x=gender_per_offer['offer_name'].unique(),
        y=gender_per_offer[(gender_per_offer['gender'] == gender)]['id_x'], name=gender))


fig.update_layout(barmode='stack', xaxis={'categoryorder':'category ascending'})
fig.show()

The share of women seems to be slightly higher for BOGO. The share of men seems to be slightly higher for discounts.

2. The age¶

Let's look at the age distribution of the persons of interest.

fig = go.Figure()

fig.add_trace(go.Histogram(x=offers_and_persons['age'], histnorm='percent', nbinsx=10))

fig.show()

We can see that most of the persons are around their 50s. Let's see if a certain offer is more popular depending on the age.

age_per_offer = offers_and_persons.groupby('offer_name')['age'].mean()

age_per_offer

offer_name
bogo_1        55.810917
bogo_2        56.965266
bogo_3        55.796725
bogo_4        55.849554
discount_1    55.185390
discount_2    54.878984
discount_3    56.087500
discount_4    56.653226
Name: age, dtype: float64

It is hard to tell by looking at the average age per offer if there is a link between the age and the offer which leads to a transaction.

3. Income¶

sum(pd.isnull(offers_and_persons['income']))

0

fig = go.Figure()

fig.add_trace(go.Histogram(x=offers_and_persons['income'], histnorm='percent'))#, nbinsx=10))

fig.show()

The income distribution does not look like to be following a Normal low (contrary to the age).

Let's see if we can see a link between the offer and the income of a person.

income_per_offer = offers_and_persons.groupby('offer_name')['income'].median()

fig = go.Figure()

fig.add_trace(go.Bar(x=income_per_offer.index, 
                     y=income_per_offer))

fig.show()

Slightly higher income will complete the BOGO offer, which can make sense.

4. Date of start membership¶

Let's transform this variable so that it translate the time lapse between the day when the person has become a member and today.

ref_date = offers_and_persons['became_member_on'].max()
ref_date

Timestamp('2018-07-26 00:00:00')

offers_and_persons['seniority'] = offers_and_persons['became_member_on'].apply(lambda x:(ref_date - x).days)

offers_and_persons.head()

/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:1: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

fig = go.Figure()
fig.add_trace(go.Histogram(x=offers_and_persons['seniority']))

fig.show()

It is clear that the more recent a client is, the more likely he will use an offer.

What about the person that complete a transaction without any offer?¶

no_offer_and_person = transcript[(~transcript['personvalue'].isin(offers_leading_to_transaction))
                                & (transcript['event'] == 'transaction')]

no_offer_and_person = no_offer_and_person[['person', 'value']].merge(profile, how='left', left_on='person', right_on='id')
no_offer_and_person

1. Gender¶

sum(pd.isnull(no_offer_and_person['gender'])) / len(no_offer_and_person)

0.1079213834893813

no_offer_and_person = no_offer_and_person[~pd.isnull(no_offer_and_person['gender'])]

fig = go.Figure()

fig.add_trace(go.Pie(values=no_offer_and_person['gender'].value_counts(normalize=True), 
                     labels=no_offer_and_person['gender'].value_counts(normalize=True).index))

fig.show()

There are more males that complete a transaction without seeing any offer.

2. Age¶

fig = go.Figure()

fig.add_trace(go.Histogram(x=no_offer_and_person['age'], histnorm='percent', nbinsx=10))

fig.show()

The age distribution looks the same as with the offers.

3. Income¶

fig = go.Figure()

fig.add_trace(go.Histogram(x=no_offer_and_person['income'], histnorm='percent', name='no offer', nbinsx=20))

fig.add_trace(go.Histogram(x=offers_and_persons['income'], histnorm='percent', name='offer', marker=dict(color='gray'), nbinsx=20))

fig.update_layout(barmode='overlay')
fig.update_traces(opacity=0.75)

fig.show()

People that don't use an offer don't have a higher income when looking at both distributions.

Still, those two distributions are quite different.

Date of membership¶

no_offer_and_person['seniority'] = no_offer_and_person['became_member_on'].apply(lambda x:(ref_date - x).days)

/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:1: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

fig = go.Figure()

fig.add_trace(go.Histogram(x=no_offer_and_person['seniority'], histnorm='percent', name='no offer', nbinsx=20))

fig.add_trace(go.Histogram(x=offers_and_persons['seniority'], histnorm='percent', name='offer', marker=dict(color='gray'), nbinsx=20))

fig.update_layout(barmode='overlay')
fig.update_traces(opacity=0.75)

fig.show()

People using an offer or not have the same distribution of date if membership.

End of data cleaning¶

As the no-offer profile is interesting to exploit, we will add an offer call 'no_offer' which will mean that for this type of person, we should not show any offer. We will thus add these profile to our main datafram offers_and_person.

no_offer_and_person['offer_name'] = 'no_offer'

no_offer_and_person.head()

/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:1: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

offers_and_persons.head()

cols_to_keep = ['person', 'offer_name', 'gender', 'age', 'seniority']

df = pd.concat([offers_and_persons[cols_to_keep], 
                no_offer_and_person[cols_to_keep]], 
                axis=0, 
                sort=False)

df['offer_name'].value_counts(normalize=True)

no_offer      0.836299
discount_1    0.030569
discount_2    0.029213
bogo_1        0.023978
bogo_2        0.019424
bogo_3        0.018951
bogo_4        0.015875
discount_3    0.015652
discount_4    0.010039
Name: offer_name, dtype: float64

We notice that when we add the no offer type, we result in an imbalance dataset. We will then exclude them for now and limit ourselves to the offers_and_persons dataset.

df = offers_and_persons[['offer_name', 'gender', 'age', 'income', 'seniority']]

df.head()

Let's one hot encode the gender column.

df = pd.concat([df.drop('gender', axis=1), 
                pd.get_dummies(df['gender'], prefix='gender', prefix_sep='_', drop_first=True)],
               axis=1, 
               sort=False)

df.head()

df['offer_name'].value_counts(normalize=True)

discount_1    0.186738
discount_2    0.178454
bogo_1        0.146472
bogo_2        0.118653
bogo_3        0.115768
bogo_4        0.096975
discount_3    0.095615
discount_4    0.061325
Name: offer_name, dtype: float64

pd.isnull(df).sum()

offer_name    0
age           0
income        0
seniority     0
gender_M      0
gender_O      0
dtype: int64

df.to_csv('clean_data.csv')

from subprocess import call
call(['python', '-m', 'nbconvert', 'explore_and_clean.ipynb'])

1

	gender	age	id	became_member_on	income
0	None	118	68be06ca386d4c31939f3a4f0e3dd783	20170212	NaN
1	F	55	0610b486422d4921ae7d2bf64640c50b	20170715	112000.0
2	None	118	38fe809add3b4fcf9315a9694bb96ff5	20180712	NaN
3	F	75	78afa995795e4d85b5d9ceeca43f5fef	20170509	100000.0
4	None	118	a03223e636434f42ac4c3df47e8bac43	20170804	NaN

	gender	age	id	became_member_on	income
1	F	55	0610b486422d4921ae7d2bf64640c50b	2017-07-15	112000.0
3	F	75	78afa995795e4d85b5d9ceeca43f5fef	2017-05-09	100000.0
5	M	68	e2127556f4f64592b11af22de27a7932	2018-04-26	70000.0
8	M	65	389bc3fa690240e798340f5a15918d5c	2018-02-09	53000.0
12	M	58	2eeac8d8feae4a8cad5a6af0499a211d	2017-11-11	51000.0

	person	event	value
0	78afa995795e4d85b5d9ceeca43f5fef	offer received	9b98b8c7a33c4b65b9aebfe6a799e6d9
1	a03223e636434f42ac4c3df47e8bac43	offer received	0b1e1539f2cc45b7b9fa7c272da2e1d7
2	e2127556f4f64592b11af22de27a7932	offer received	2906b810c7d4411798c6938adc9daaa5
3	8ec6ce2a7e7949b1bf142def7d0e0586	offer received	fafdcd668e3743c1bb461111dcafc2a4
4	68617ca6246f4fbc85e91a2a49552598	offer received	4d5c57ea9a6940dd891ad53e9dbe8da0

	person	event	value	personvalue
0	78afa995795e4d85b5d9ceeca43f5fef	offer received	9b98b8c7a33c4b65b9aebfe6a799e6d9	78afa995795e4d85b5d9ceeca43f5fef-9b98b8c7a33c4...
1	a03223e636434f42ac4c3df47e8bac43	offer received	0b1e1539f2cc45b7b9fa7c272da2e1d7	a03223e636434f42ac4c3df47e8bac43-0b1e1539f2cc4...
2	e2127556f4f64592b11af22de27a7932	offer received	2906b810c7d4411798c6938adc9daaa5	e2127556f4f64592b11af22de27a7932-2906b810c7d44...
3	8ec6ce2a7e7949b1bf142def7d0e0586	offer received	fafdcd668e3743c1bb461111dcafc2a4	8ec6ce2a7e7949b1bf142def7d0e0586-fafdcd668e374...
4	68617ca6246f4fbc85e91a2a49552598	offer received	4d5c57ea9a6940dd891ad53e9dbe8da0	68617ca6246f4fbc85e91a2a49552598-4d5c57ea9a694...

	reward	channels	difficulty	duration	offer_type	id	share_of_transactions
6	2	[web, email, mobile, social]	10	10	discount	fafdcd668e3743c1bb461111dcafc2a4	0.189977
5	3	[web, email, mobile, social]	7	7	discount	2298d6c36e964ae4a3e7e9706d1fb8c2	0.181706
8	5	[web, email, mobile, social]	5	5	bogo	f19421c1d4aa40978ebb69ca19b0e20d	0.150550
1	10	[web, email, mobile, social]	10	5	bogo	4d5c57ea9a6940dd891ad53e9dbe8da0	0.116646
0	10	[email, mobile, social]	10	7	bogo	ae264e3637204a6fb9bb56bc8210ddfd	0.112948
3	5	[web, email, mobile]	5	7	bogo	9b98b8c7a33c4b65b9aebfe6a799e6d9	0.095546
9	2	[web, email, mobile]	10	7	discount	2906b810c7d4411798c6938adc9daaa5	0.094494
4	5	[web, email]	20	10	discount	0b1e1539f2cc45b7b9fa7c272da2e1d7	0.058133
2	0	[web, email, mobile]	0	4	informational	3f207df678b143eea3cee63160fa8bed	NaN
7	0	[email, mobile, social]	0	3	informational	5a8bc65990b245e5a138643cd4eb9837	NaN

	person	value
12658	9fa9ae8f57894cc9a3b8a9bbe0fc1b2f	2906b810c7d4411798c6938adc9daaa5
12679	629fc02d56414d91bca360decdfa9288	9b98b8c7a33c4b65b9aebfe6a799e6d9
12692	676506bad68e4161b9bbaffeb039626b	ae264e3637204a6fb9bb56bc8210ddfd
12697	8f7dd3b2afe14c078eb4f6e6fe4ba97d	4d5c57ea9a6940dd891ad53e9dbe8da0
12717	227f2d69e46a4899b70d48182822cff6	4d5c57ea9a6940dd891ad53e9dbe8da0

	person	value_x	value_y
0	9fa9ae8f57894cc9a3b8a9bbe0fc1b2f	2906b810c7d4411798c6938adc9daaa5	34.56
1	629fc02d56414d91bca360decdfa9288	9b98b8c7a33c4b65b9aebfe6a799e6d9	33.9
2	676506bad68e4161b9bbaffeb039626b	ae264e3637204a6fb9bb56bc8210ddfd	18.01
3	8f7dd3b2afe14c078eb4f6e6fe4ba97d	4d5c57ea9a6940dd891ad53e9dbe8da0	19.11
4	227f2d69e46a4899b70d48182822cff6	4d5c57ea9a6940dd891ad53e9dbe8da0	28.39

	person	offer_name
0	9fa9ae8f57894cc9a3b8a9bbe0fc1b2f	discount_3
1	629fc02d56414d91bca360decdfa9288	bogo_4
2	676506bad68e4161b9bbaffeb039626b	bogo_3
3	8f7dd3b2afe14c078eb4f6e6fe4ba97d	bogo_2
4	227f2d69e46a4899b70d48182822cff6	bogo_2
...	...	...
25233	0c027f5f34dd4b9eba0a25785c611273	discount_2
25234	a6f84f4e976f44508c358cc9aba6d2b3	discount_2
25235	b895c57e8cd047a8872ce02aa54759d6	discount_1
25236	8431c16f8e1d440880db371a68f82dd0	discount_1
25237	24f56b5e1849462093931b164eb803b5	discount_1

	person	offer_name	gender	age	id	became_member_on	income
0	9fa9ae8f57894cc9a3b8a9bbe0fc1b2f	discount_3	M	42.0	9fa9ae8f57894cc9a3b8a9bbe0fc1b2f	2016-01-17	96000.0
1	629fc02d56414d91bca360decdfa9288	bogo_4	M	52.0	629fc02d56414d91bca360decdfa9288	2018-06-05	72000.0
2	676506bad68e4161b9bbaffeb039626b	bogo_3	M	37.0	676506bad68e4161b9bbaffeb039626b	2017-05-15	92000.0
3	8f7dd3b2afe14c078eb4f6e6fe4ba97d	bogo_2	M	48.0	8f7dd3b2afe14c078eb4f6e6fe4ba97d	2015-09-03	62000.0
4	227f2d69e46a4899b70d48182822cff6	bogo_2	M	51.0	227f2d69e46a4899b70d48182822cff6	2018-05-12	100000.0
...	...	...	...	...	...	...	...
25233	0c027f5f34dd4b9eba0a25785c611273	discount_2	M	56.0	0c027f5f34dd4b9eba0a25785c611273	2017-10-24	61000.0
25234	a6f84f4e976f44508c358cc9aba6d2b3	discount_2	NaN	NaN	NaN	NaT	NaN
25235	b895c57e8cd047a8872ce02aa54759d6	discount_1	NaN	NaN	NaN	NaT	NaN
25236	8431c16f8e1d440880db371a68f82dd0	discount_1	M	39.0	8431c16f8e1d440880db371a68f82dd0	2018-06-27	39000.0
25237	24f56b5e1849462093931b164eb803b5	discount_1	F	48.0	24f56b5e1849462093931b164eb803b5	2017-12-28	80000.0

	offer_name	gender	id_x	id_y
0	bogo_1	F	0.463421	3554
1	bogo_1	M	0.519415	3554
2	bogo_1	O	0.017164	3554
3	bogo_2	F	0.534213	2879
4	bogo_2	M	0.451198	2879

	person	value	gender	age	id	became_member_on	income
0	02c083884c7d45b39cc68e1314fec56c	0.83	F	20.0	02c083884c7d45b39cc68e1314fec56c	2016-07-11	30000.0
1	9fa9ae8f57894cc9a3b8a9bbe0fc1b2f	34.56	M	42.0	9fa9ae8f57894cc9a3b8a9bbe0fc1b2f	2016-01-17	96000.0
2	54890f68699049c2a04d415abc25e717	13.23	M	36.0	54890f68699049c2a04d415abc25e717	2017-12-28	56000.0
3	b2f1cd155b864803ad8334cdf13c4bd2	19.51	F	55.0	b2f1cd155b864803ad8334cdf13c4bd2	2017-10-16	94000.0
4	fe97aa22dd3e48c8b143116a8403dd52	18.97	F	39.0	fe97aa22dd3e48c8b143116a8403dd52	2017-12-17	67000.0
...	...	...	...	...	...	...	...
138948	b3a1272bc9904337b331bf348c3e8c17	1.59	M	66.0	b3a1272bc9904337b331bf348c3e8c17	2018-01-01	47000.0
138949	68213b08d99a4ae1b0dcb72aebd9aa35	9.53	M	52.0	68213b08d99a4ae1b0dcb72aebd9aa35	2018-04-08	62000.0
138950	a00058cf10334a308c68e7631c529907	3.61	F	63.0	a00058cf10334a308c68e7631c529907	2013-09-22	52000.0
138951	76ddbd6576844afe811f1a3c0fbb5bec	3.53	M	57.0	76ddbd6576844afe811f1a3c0fbb5bec	2016-07-09	40000.0
138952	c02b10e8752c4d8e9b73f918558531f7	4.05	NaN	NaN	NaN	NaT	NaN